Customer Decisioning Glossary

Common terms used by customer decisioning practitioners

Customer Decisioning Glossary

  • Arbitration: Using a formula incorporating dynamic elements such as relevance and value to compute priority.

  • Auditability: Audit trail capabilities that ensures the history for each decision and it’s elements at an individual customer level can be retrieved and explained.

  • Contextual Arbitration: Using the real-time customer context (Eg. Looking at a product page, transacting, complaining, churning, etc) within arbitration. This is critical for customer relevance.

  • Eligibility: Rules to determine if a customer qualifies for a specific message. Contact rules to manage and enforce communication limits with customers that can be applied at the channel or action.

  • Individual Level Decisions: AKA 1:1 customer engagement that allows decisioning for a unique customer not a segment of customers. Takes into consideration the unique behaviors, interests, and intents of each customer to drive highly personalized customer experiences that can adapt instantly to changes in customer intent and event data.

  • In-Session Context Customer Signals: Using the context of what a customer is doing on a page, call, action responses, etc to re-decision what is the next best action.

  • In-Session self-learning models outside of decisioning system but with real-time AI: Connectors to call out to external AI services such as Google AI or AWS SageMaker, and then integrate the results on-demand at the point of a decision being made.

  • In-Session self-learning engagement models within decisioning system: Self-learning AI such as Adaptive AI or Reinforcement Learning that predicts likelihood to engage that can be executed at the point of decisioning.

  • Self-Learning AI: Systems adjust their behavior in real time based on changing data or performance metrics, rather than relying on fixed models trained once and deployed. Please see below the three types used in customer decisioning.

    • Adaptive Bayesian: Bayesian approaches treat uncertainty as probability distributions over possible values or models, then update those distributions as new data arrives using Bayes' theorem. In decisioning, this means maintaining confidence estimates about what works (e.g., "80% confident email converts better than SMS for this customer"), then revising those estimates each time you observe an outcome, allowing principled decisions under uncertainty. Decision Transparency: High

      Adaptive Gradient Boosting: Gradient boosting builds an ensemble of weak models (usually decision trees) sequentially, where each new model corrects the errors of the previous ones by focusing on instances the ensemble got wrong. Adaptive gradient boosting adds real-time adjustment—monitoring prediction errors and retaining or recalibrating the ensemble as new customer data arrives, keeping the model responsive to shifts in behavior patterns. Decision Transparency: Low > Medium depending on requirements

    • Reinforcement Learning: Reinforcement learning is a machine learning approach where an agent learns by interacting with an environment: it takes an action, receives a reward (or penalty), and adjusts its strategy to maximize cumulative reward over time. The agent uses algorithms like Q-learning or Thompson Sampling to balance exploration (trying new actions to gather information) and exploitation (repeating actions that worked well). Unlike supervised learning, which learns from labeled examples, RL learns from trial-and-error feedback, gradually building a policy—a mapping of situations to optimal actions. Effective but very limited transparence into how a decision was made. Decision Transparency: Low

  • In-Session self-learning conversion models within decisioning system: Self-learning AI that predicts likelihood for a customer to convert that can be executed at the point of decisioning.

  • In-Session Contextual Arbitration: The ability to instantly re-run a decisioning (AKA re-decisioning) as an individual customer’s context changes to maximize the potential for positive outcomes.

  • Journey Orchestration: The process of managing and coordinating the series of interactions between a customer and an organization that occur as the customer pursues a specific goal. It involves understanding the customer's needs and preferences, and using this information to guide the customer through their journey.

  • Live Event Streaming Data: The capability to monitor and use streams of events such as transactions, calls, IVRs and digital channel signals.

  • Next Best Action: A customer engagement strategy to present customers with the most relevant and/or profitable message.

  • Off-Line Model Scoring Real-Time Execution: Real-time execution of off-line models in decisioning. Scores produced by data scientists using their preferred tools and then imported for operational use via PMML or MOJO, allowing models from SAS, SPSS, R, Python, etc to be imported and executed at the point of decisioning to produce the latest score with every customer interaction.

  • Off-Line Model Scoring: (BATCH EXECUTION) Predictive models that are built and scored outside of the core decisioning solution. The scores are then passed into decisioning on a regular basis to feed decisions.

  • Prioritization: The process of ranking hierarchically actions based on certain criteria such as relevance and/or profit.

  • Segment Level Decisions: Decisions made for a segment e.g. All customers would receive the same next best action within the segment. (no matter how small or large)

  • Segmentation: The ability to create or import and report on segments of customers.

  • Simulation / Forecasting: Running simulation tests to understand the financial and CX impact of adding, changing or up weighting actions to help make important business decisions.